Academic Journal
Προβολή σε EDS
Συνδεδεμένο Πλήρες Κείμενο

Domain generalisation challenges in breast cancer molecular classification using foundation models: a cross-cohort exploratory study.

Λεπτομέρειες βιβλιογραφικής εγγραφής
Τίτλος: Domain generalisation challenges in breast cancer molecular classification using foundation models: a cross-cohort exploratory study.
Συγγραφείς: Fernandez-Romero J; Department of Computer Languages and Systems, ETSII, University of Seville, Av. Reina Mercedes s/n, 41012, Seville, Andalusia, Spain., Ramos-Berciano P; Department of Computer Languages and Systems, ETSII, University of Seville, Av. Reina Mercedes s/n, 41012, Seville, Andalusia, Spain., Perez-Perez M; Department of Normal and Pathological Cytology and Histology, Faculty of Medicine, University of Seville, Av. Doctor Fadriani s/n, 41009, Seville, Andalusia, Spain., Benavides D; Department of Computer Languages and Systems, ETSII, University of Seville, Av. Reina Mercedes s/n, 41012, Seville, Andalusia, Spain., Robles-Frias A; UGC Pathology, Hospital Universitario Virgen de Valme, Ctra. de Cádiz Km. 548, 41004, Seville, Andalusia, Spain., Garcia-Gutierrez J; Department of Computer Languages and Systems, ETSII, University of Seville, Av. Reina Mercedes s/n, 41012, Seville, Andalusia, Spain. jorgarcia@us.es., Macias-Garcia L; Department of Normal and Pathological Cytology and Histology, Faculty of Medicine, University of Seville, Av. Doctor Fadriani s/n, 41009, Seville, Andalusia, Spain.
Πηγή: Medical & biological engineering & computing [Med Biol Eng Comput] 2026 Jun; Vol. 64 (6), pp. 2321-2331. Date of Electronic Publication: 2026 May 11.
Τύπος έκδοσης: Journal Article
Γλώσσα: English
Στοιχεία περιοδικού: Publisher: Springer Country of Publication: United States NLM ID: 7704869 Publication Model: Print-Electronic Cited Medium: Internet ISSN: 1741-0444 (Electronic) Linking ISSN: 01400118 NLM ISO Abbreviation: Med Biol Eng Comput Subsets: MEDLINE
Imprint Name(s): Publication: New York, NY : Springer
Original Publication: Stevenage, Eng., Peregrinus.
Ιατρικοί όροι (MeSH): Breast Neoplasms*/classification , Breast Neoplasms*/genetics , Breast Neoplasms*/metabolism , Breast Neoplasms*/pathology , Classification Algorithms* , Multiple-Instance Learning Algorithms*, Biomarkers, Tumor/metabolism ; Erb-b2 Receptor Tyrosine Kinases/metabolism ; Female ; Humans ; Cohort Studies ; Immunohistochemistry
Περίληψη: Molecular classification guides breast cancer treatment, but PAM50 and immunohistochemistry (IHC) remain costly and unavailable in many settings. Foundation models (FMs) combined with multiple instance learning (MIL) show promise for predicting molecular subtypes from haematoxylin-and-eosin-stained slides, yet most studies report only internal validation. This study evaluates FMs with MIL across cohorts and identifies factors associated with domain-induced performance degradation. We evaluate 13 FMs and 3 complementary MIL architectures for PAM50 subtyping and IHC biomarker prediction using cross-validation on TCGA-BRCA ([Formula: see text]) and external validation on CPTAC-BRCA ([Formula: see text]). Virchow v2 achieves the best overall performance but exhibits severe degradation upon external validation, consistent across all three MIL architectures especially for HER2-enriched and Normal-like PAM50 subtypes and HER2-positive IHC prediction. Four hypothesised domain shift factors are quantified through exploratory regression analysis to explain relative performance drop (RPD). Staining variability, feature space divergence and morphological separability reach significance in univariate analysis, whilst prevalence shift does not. Staining variability and feature space divergence as covariate-level factors jointly account for 80.0% of RPD variance in the most parsimonious multivariate model ([Formula: see text], [Formula: see text]). Although based on a limited number of class-level observations and therefore exploratory in nature, these findings highlight the need for domain generalisation strategies targeting covariate shift, even when specialised FMs are used as feature encoders.
(© 2026. The Author(s).)
Competing Interests: Declarations. Conflict of interest: The authors have no competing interests to declare that are relevant to the content of this article.
References: Zhang Y, Ji Y, Liu S et al (2025) Global burden of female breast cancer: new estimates in 2022, temporal trend and future projections up to 2050 based on the latest release from GLOBOCAN. J Natl Cancer Center 5(3):287–296. https://doi.org/10.1016/j.jncc.2025.02.002. (PMID: 10.1016/j.jncc.2025.02.002)
Wallden B, Storhoff J, Nielsen T, et al (2015) Development and verification of the PAM50-based Prosigna breast cancer gene signature assay. BMC Med Genomics 8(54).
Ziegengeist JL, Tan AR (2025) A clinical review of subcutaneous trastuzumab and the fixed-dose combination of pertuzumab and trastuzumab for subcutaneous injection in the treatment of HER2-positive breast cancer. Clin Breast Cancer 25(2):124–132. https://doi.org/10.1016/j.clbc.2024.10.005. (PMID: 10.1016/j.clbc.2024.10.005)
Tafavvoghi M, Sildnes A, Rakaee M et al (2025) Deep learning-based classification of breast cancer molecular subtypes from H&E whole-slide images. J Pathology Inf 16:100410. https://doi.org/10.1016/j.jpi.2024.100410. (PMID: 10.1016/j.jpi.2024.100410)
Niyas S, Bygari R, Naik R et al (2023) Automated molecular subtyping of breast carcinoma using deep learning techniques. IEEE J Trans Eng Health Med 11:161–169. https://doi.org/10.1109/JTEHM.2023.3241613. (PMID: 10.1109/JTEHM.2023.3241613)
Huang J, Li G, Kan S et al (2025) An efficient framework based on large foundation model for cervical cytopathology whole slide image screening. Biomed Signal Process Control 107:107859. https://doi.org/10.1016/j.bspc.2025.107859. (PMID: 10.1016/j.bspc.2025.107859)
Ma J, Xu Y, Zhou F et al (2025) PathBench: A comprehensive comparison benchmark for pathology foundation models towards precision oncology. arxiv:2505.20202.
Valieris R, Martins L, Defelicibus A et al (2024) Weakly-supervised deep learning models enable HER2-low prediction from H&E stained slides. Breast Cancer Res 26:124. https://doi.org/10.1186/s13058-024-01863-0. (PMID: 10.1186/s13058-024-01863-03916059311331614)
Antamis T, Drosou A, Vafeiadis T et al (2024) Interpretability of deep neural networks: A review of methods, classification and hardware. Neurocomputing 601:128204. https://doi.org/10.1016/j.neucom.2024.128204. (PMID: 10.1016/j.neucom.2024.128204)
Dolezal JM, Wolk R, Hieromnimon HM, et al (2023) Deep learning generates synthetic cancer histology for explainability and education. NPJ Precis Onc 7(23). https://doi.org/10.1038/s41698-023-00399-4.
Lu MY, Williamson DF, Chen T et al (2021) Data-efficient and weakly supervised computational pathology on whole-slide images. Nat Biomed Eng 5(6):555–570. https://doi.org/10.1038/s41551-020-00682-w. (PMID: 10.1038/s41551-020-00682-w336495648711640)
Shi J, Sun D, Jiang Z et al (2025) Weakly supervised multi-modal contrastive learning framework for predicting the her2 scores in breast cancer. Comput Med Imaging Graph 121:102502. https://doi.org/10.1016/j.compmedimag.2025.102502. (PMID: 10.1016/j.compmedimag.2025.10250239919535)
Pérez-Núñez J, Rodríguez C, Vásquez-Serpa L et al (2024) The challenge of deep learning for the prevention and automatic diagnosis of breast cancer: a systematic review. Diagnostics (Basel) 14(24):2896. https://doi.org/10.3390/diagnostics14242896. (PMID: 10.3390/diagnostics142428963976725711675111)
Bisson T, Franz M, Kiehl TR et al (2025) A high-precision hierarchical registration approach for stain- and scanner-independent colocalization on whole slide images in histopathology. Health Inf Sci Syst 13:38. https://doi.org/10.1007/s13755-025-00353-7. (PMID: 10.1007/s13755-025-00353-74041651512102413)
Godau P, Kalinowski P, Christodoulou E et al (2025) Navigating prevalence shifts in image analysis algorithm deployment. Med Image Anal 102:103504. https://doi.org/10.1016/j.media.2025.103504. (PMID: 10.1016/j.media.2025.10350440020420)
Gupta E, Gupta V (2025) Margin-aware optimized contrastive learning for enhanced self-supervised histopathological image classification. Health Inf Sci Syst 13:2. https://doi.org/10.1007/s13755-024-00316-4. (PMID: 10.1007/s13755-024-00316-439619405)
Lingle W, Erickson BJ, Zuley ML et al (2016) The cancer genome atlas breast invasive carcinoma collection (TCGA-BRCA) (Version 3) [Data set]. Cancer Imaging Arch. https://doi.org/10.7937/K9/TCIA.2016.AB2NAZRP. (PMID: 10.7937/K9/TCIA.2016.AB2NAZRP)
Lindgren CM, Adams DW, Kimball B et al (2021). Simplified and unified access to cancer proteogenomic data. https://doi.org/10.1021/acs.jproteome.0c00919.
Thennavan A, Beca F, Xia Y et al (2021) Molecular analysis of TCGA breast cancer histologic types. Cell Genomics 1(3):100067. https://doi.org/10.1016/j.xgen.2021.100067. (PMID: 10.1016/j.xgen.2021.10006735465400)
Krug K, Jaehnig EJ, Satpathy S et al (2020) Proteogenomic landscape of breast cancer tumorigenesis and targeted therapy. Cell 183(5):1436–145631. https://doi.org/10.1016/j.cell.2020.10.036. (PMID: 10.1016/j.cell.2020.10.036332120108077737)
Brussee S, Valkema PA, Weijer JA, Doeleman T, Schrader AM, Kers J (2025) Pathbench-mil: A comprehensive automl and benchmarking framework for multiple instance learning in histopathology. arXiv preprint arXiv:2512.17517.
Garcia-Gutierrez J (2025) CLAiMem-ALL. https://github.com/BIGS-investigacion/CLAiMem-ALL.git.
Garcia-Gutierrez J (2025) CLAiMem-ALL. https://github.com/BIGS-Investigacion/PathBench-MIL.git.
Pedregosa F, Varoquaux G, Gramfort A et al (2011) Scikit-learn: machine learning in python. J Mach Learn Res 12:2825–2830.
Shao Z, Bian H, Chen Y et al (2021) Transmil: Transformer based correlated multiple instance learning for whole slide image classification. Adv Neural Inf Process Syst 34:2136–2147.
Li B, Li Y, Eliceiri KW (2021) Dual-stream multiple instance learning network for whole slide image classification with self-supervised contrastive learning. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp 14318–14328.
Akiba T, Sano S, Yanase T, Ohta T, Koyama M (2019) Optuna: a next-generation hyperparameter optimization framework. In: The 25th ACM SIGKDD international conference on knowledge discovery & data mining, pp 2623–2631.
Macenko M, Niethammer M, Marron JS et al (2009) A method for normalizing histology slides for quantitative analysis. In: 2009 IEEE international symposium on biomedical imaging: from nano to macro. IEEE, pp 1107–1110. https://doi.org/10.1109/ISBI.2009.5193250.
Dolezal JM, Kochanny S, Dyer E et al (2024) Slideflow: deep learning for digital histopathology with real-time whole-slide visualization. BMC Bioinform 25(1):134. https://doi.org/10.1186/s12859-024-05758-x. (PMID: 10.1186/s12859-024-05758-x)
Ginter PS, Idress R, D’Alfonso TM, Fineberg S, Jaffer S, Sattar AK, Chagpar A, Wilson P, Harigopal M (2021) Histologic grading of breast carcinoma: a multi-institution study of interobserver variation using virtual microscopy. Mod Pathol 34(4):701–709. https://doi.org/10.1038/s41379-020-00698-2. (PMID: 10.1038/s41379-020-00698-233077923)
Landis JR, Koch GG (1977) The measurement of observer agreement for categorical data. Biometrics 33(1):159–174. (PMID: 10.2307/2529310843571)
Shamai G, Schley R, Cretu A et al (2024) Clinical utility of receptor status prediction in breast cancer and misdiagnosis identification using deep learning on hematoxylin and eosin-stained slides. Commun Med 4(1):276. https://doi.org/10.1038/s43856-024-00695-5. (PMID: 10.1038/s43856-024-00695-53970686111661999)
Jang W, Lee J, Park K et al (2024) Molecular classification of breast cancer using weakly supervised learning. Cancer Res Treat 57(1):116–125. https://doi.org/10.4143/crt.2024.113. (PMID: 10.4143/crt.2024.1133893801011729310)
Jahanifar M, Raza M, Xu K et al (2025) Domain generalization in computational pathology: Survey and guidelines. ACM Comput Surv 57(11). https://doi.org/10.1145/3724391.
Tellez D, Litjens G, Bándi P et al (2019) Quantifying the effects of data augmentation and stain color normalization in convolutional neural networks for computational pathology. Med Image Anal 58:101544. https://doi.org/10.1016/j.media.2019.101544. (PMID: 10.1016/j.media.2019.10154431466046)
Hu EJ, Shen Y, Wallis P et al (2022) LoRA: Low-rank adaptation of large language models. In: International conference on learning representations.
Li Z, Ren K, Jiang X, Shen Y, Zhang H, Li D (2023) SIMPLE: Specialized model-sample matching for domain generalization. In: The eleventh international conference on learning representations.
Lee G, Jang W, Kim J et al. Domain generalization using large pretrained models with mixture-of-adapters. https://doi.org/10.1109/WACV61041.2025.00801.
Contributed Indexing: Keywords: Breast cancer; Computational pathology; Domain shift; External validation; Foundation models
Substance Nomenclature: 0 (Biomarkers, Tumor)
EC 2.7.10.1 (Erb-b2 Receptor Tyrosine Kinases)
Entry Date(s): Date Created: 20260511 Date Completed: 20260615 Latest Revision: 20260726
Update Code: 20260726
PubMed Central ID: PMC13269319
DOI: 10.1007/s11517-026-03590-4
PMID: 42113320
Βάση Δεδομένων: MEDLINE
Περιγραφή
ISSN:1741-0444
DOI:10.1007/s11517-026-03590-4